FuzzyAI applies fuzz-testing ideas to large language model APIs by running documented attack examples against configured model targets. Developed by CyberArk's security research team, the Apache-2.0 project currently documents jailbreak-oriented fuzzing plus examples for ManyShot, Taxonomy, and ArtPrompt-style attacks across OpenAI, Anthropic, Ollama/local models, and custom REST APIs.
The framework operates against documented provider examples rather than a guaranteed universal model matrix: OpenAI, Anthropic, Ollama/local models, and custom REST API targets are visible in the current README. Teams that need governance-grade evidence should wrap the CLI output with their own severity taxonomy, storage, and remediation workflow. This evidence-based approach helps security teams quantify LLM risk rather than relying on qualitative assessments of model safety.
With 1.4K+ GitHub stars and CyberArk stewardship, FuzzyAI fills a useful gap for teams that want a repeatable starting point for LLM security testing without building every fuzzing prompt from scratch. The repository was active in the source check, but the latest push observed in this pass was 2026-02-06, so production users should review current issue activity and code paths before standardizing on it.